-
Notifications
You must be signed in to change notification settings - Fork 0
/
save_select_then_predict.py
321 lines (242 loc) · 12.4 KB
/
save_select_then_predict.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
# 1. bert predictive resultes -- on In domain / ood1 / ood2
# 2. faithful: for both top / contigious -- on In domain / ood1 / ood2
# 3. FRESH results
# 4. kuma results (another script)
# 5. domain similarity between: In domain / ood1 / ood2
# 6. rationale similarity between: In domain / ood1 / ood2
# 7. datasets metadata: train/test/ size, time span, label distribution
import pandas as pd
import json
import csv
import config.cfg
import os
import argparse
import fnmatch
import numpy as np
parser = argparse.ArgumentParser()
parser.add_argument(
"--dataset",
type = str,
help = "select dataset / task",
default = "complain",
)
parser.add_argument(
"--combine_all_tasks",
help='combine all',
action='store_true',
default=False
)
parser.add_argument(
'--get_all_seeds_for_predictive',
help='get all seeds results for bert prediction',
action='store_true',
default=False
)
parser.add_argument(
'--save_for_bert',
help='decide which parts are in need',
action='store_true',
default=False
)
parser.add_argument(
'--save_for_fresh',
help='decide which parts are in need',
action='store_true',
default=False
)
parser.add_argument(
'--save_for_kuma_lstm',
help='decide which parts are in need',
action='store_true',
default=False
)
parser.add_argument(
'--save_posthoc',
help='decide which parts are in need',
action='store_true',
default=False
)
parser.add_argument(
'--save_data_stat',
help='decide which parts are in need',
action='store_true',
default=False
)
parser.add_argument(
'--save_accuracy_version',
help='decide which parts are in need',
action='store_true',
default=False
)
### 最完整的版本，直接提json的结果 最近期的
args = parser.parse_args()
datasets_dir = 'saved_everything/' + str(args.dataset)
os.makedirs(datasets_dir, exist_ok = True)
if args.combine_all_tasks:
task_list = ['agnews', 'xfact', 'factcheck', 'AmazDigiMu', 'AmazPantry', 'yelp']
result_list = []
for task in task_list:
result = pd.read_csv('saved_everything/' + str(task) + '/selective_results.csv')
result['Task'] = str(task)
result_list.append(result)
final = pd.concat(result_list)
print(final)
print(final.dtypes)
final.to_csv('./saved_everything/all_tasks_all_selective.csv')
exit()
if args.save_accuracy_version:
select_columns = ['mean-acc', 'std-acc']
else:
select_columns = ['mean-f1', 'std-f1']
######################## 1. bert predictive resultes -- on In domain / ood1 / ood2
InDomain = pd.read_json('./models/'+str(args.dataset)+'/bert_predictive_performances.json')
Full_data = pd.read_json('./models/'+str(args.dataset)+'_full/bert_predictive_performances.json')
path = os.path.join('./models/', str(args.dataset),'bert_predictive_performances-OOD-' + str(args.dataset) + '_ood1.json')
OOD1 = pd.read_json(path)
path = os.path.join('./models/', str(args.dataset),'bert_predictive_performances-OOD-' + str(args.dataset) + '_ood2.json')
OOD2 = pd.read_json(path)
Full_data = Full_data[select_columns].iloc[0]
InDomain = InDomain[select_columns].iloc[0]
OOD1 = OOD1[select_columns].iloc[0]
OOD2 = OOD2[select_columns].iloc[0]
Full_data['Domain'] = 'Full'
InDomain['Domain'] = 'SynD'
OOD1['Domain'] = 'AsyD1'
OOD2['Domain'] = 'AsyD2'
bert_result = pd.concat([Full_data, InDomain, OOD1, OOD2], ignore_index=False, axis=1).T
cols = bert_result.columns.tolist()
cols = cols[-1:] + cols[:-1]
bert_result = bert_result[cols]
if args.save_accuracy_version:
bert_result = bert_result.reset_index()[['Domain', 'mean-acc', 'std-acc']]
bert_result = bert_result.rename(columns={"mean-acc":"BERT ACC", "std-acc":"BERT ACC"})
else:
bert_result = bert_result.reset_index()[['Domain', 'mean-f1', 'std-f1']]
bert_result = bert_result.rename(columns={"mean-f1":"BERT F1", "std-f1":"BERT std"})
bert_result['BERT F1'] = bert_result['BERT F1'].astype(float, errors = 'raise')
bert_result['BERT std'] = bert_result['BERT std'].astype(float, errors = 'raise')
#bert_result.to_csv('./saved_everything/'+str(args.dataset)+'/bert_predictive.csv') # less column
####################################################################################
#####################################################################################
########################### 3. FRESH results of top scaled attention
import os.path
if args.dataset == 'AmazDigiMu':
fresh_OOD1 = pd.read_json('FRESH_classifiers/AmazDigiMu/topk/scaled attention_bert_predictive_performances-OOD-AmazDigiMu_ood1.json')
fresh_OOD2 = pd.read_json('FRESH_classifiers/AmazDigiMu/topk/scaled attention_bert_predictive_performances-OOD-AmazDigiMu_ood2.json')
fresh_full_data = pd.read_json('./FRESH_classifiers/AmazDigiMu_full/topk/scaled attention_bert_predictive_performances.json')
fresh_InDomain = pd.read_json('./FRESH_classifiers/AmazDigiMu/topk/scaled attention_bert_predictive_performances.json')
elif args.dataset == 'AmazPantry':
fresh_OOD1 = pd.read_json('FRESH_classifiers/AmazPantry/topk/scaled attention_bert_predictive_performances-OOD-AmazPantry_ood1.json')
fresh_OOD2 = pd.read_json('FRESH_classifiers/AmazPantry/topk/scaled attention_bert_predictive_performances-OOD-AmazPantry_ood2.json')
fresh_full_data = pd.read_json('./FRESH_classifiers/AmazPantry_full/topk/scaled attention_bert_predictive_performances.json')
fresh_InDomain = pd.read_json('./FRESH_classifiers/AmazPantry/topk/scaled attention_bert_predictive_performances.json')
else:
fresh_OOD1_path = os.path.join('FRESH_classifiers', str(args.dataset), 'topk/', 'scaled attention_bert_predictive_performances-OOD-' + str(args.dataset) + '_ood1.json')
fresh_OOD2_path = os.path.join('FRESH_classifiers', str(args.dataset), 'topk/', 'scaled attention_bert_predictive_performances-OOD-' + str(args.dataset) + '_ood2.json')
try:
file_exists = os.path.exists(fresh_OOD1_path)
except:
fresh_OOD1_path = os.path.join('FRESH_classifiers', str(args.dataset), 'topk/', 'scaled_attention_bert_predictive_performances-OOD-' + str(args.dataset) + '_ood1.json')
fresh_OOD2_path = os.path.join('FRESH_classifiers', str(args.dataset), 'topk/', 'scaled_attention_bert_predictive_performances-OOD-' + str(args.dataset) + '_ood2.json')
file_exists = os.path.exists(fresh_OOD1_path)
print('fresh ood path ')
print(fresh_OOD1_path)
fresh_OOD1 = pd.read_json(fresh_OOD1_path)
fresh_OOD2 = pd.read_json(fresh_OOD2_path)
fresh_full_data = pd.read_json(
'./FRESH_classifiers/'+str(args.dataset)+'_full/topk/scaled attention_bert_predictive_performances.json')
fresh_InDomain = pd.read_json(
'./FRESH_classifiers/' + str(args.dataset) + '/topk/scaled attention_bert_predictive_performances.json')
fresh_full_data = fresh_full_data[select_columns].iloc[1]
fresh_full_data['Domain'] = 'Full size'
fresh_InDomain = fresh_InDomain[select_columns].iloc[1]
fresh_InDomain['Domain'] = 'SynD'
fresh_OOD1 = fresh_OOD1[select_columns].iloc[1]
fresh_OOD1['Domain'] = 'AsyD1'
fresh_OOD2 = fresh_OOD2[select_columns].iloc[1]
fresh_OOD2['Domain'] = 'AsyD2'
fresh_result = pd.concat([fresh_full_data, fresh_InDomain, fresh_OOD1, fresh_OOD2], axis=1, ignore_index=False).T.reset_index()[select_columns]
if args.save_accuracy_version:
fresh_result = fresh_result.rename(columns={"mean-acc":"FRESH ACC", "std-acc":"FRESH std"})
else:
fresh_result = fresh_result.rename(columns={"mean-f1":"FRESH F1", "std-f1":"FRESH std"})
fresh_result['FRESH F1'] = fresh_result['FRESH F1'].astype(float, errors = 'raise')
fresh_result['FRESH std'] = fresh_result['FRESH std'].astype(float, errors = 'raise')
#fresh_result.to_csv('./saved_everything/'+str(args.dataset)+'/bert_predictive.csv')
#################################### KUMA AND LSTM ############################################################
## get KUMA of FULL / IN D / OOD1 / OOD2
kuma_FullData = pd.read_json('./kuma_model/'+str(args.dataset)+'_full/kuma-bert_predictive_performances.json')
# pd.options.display.max_columns = None
# print(' KUMA FULL ')
# print(kuma_FullData)
# kuma_InDomain_path = os.path.join('kuma_model',str(args.dataset),'/kuma-bert_predictive_performances.json')
# print(' KUMA kuma_InDomain_path ')
# print(kuma_InDomain_path)
kuma_InDomain = pd.read_json('./kuma_model/'+str(args.dataset)+'/kuma-bert_predictive_performances.json')
kuma_OOD1 = pd.read_json('./kuma_model/'+str(args.dataset)+'/kuma-bert_predictive_performances-OOD-' + str(args.dataset) + '_ood1.json')
kuma_OOD2 = pd.read_json('./kuma_model/'+str(args.dataset)+'/kuma-bert_predictive_performances-OOD-' + str(args.dataset) + '_ood2.json')
print(' KUMA OOD1 ')
print(kuma_OOD1)
LSTM_FullData = pd.read_json(
'./LSTM_model/' + str(args.dataset) + '_full/full_lstm-bert_predictive_performances.json')
LSTM_InDomain = pd.read_json('./LSTM_model/' + str(args.dataset) + '/full_lstm-bert_predictive_performances.json')
LSTM_OOD1 = pd.read_json('./LSTM_model/' + str(args.dataset) + '/full_lstm-bert_predictive_performances-OOD-' + str(
args.dataset) + '_ood1.json')
LSTM_OOD2 = pd.read_json('./LSTM_model/' + str(args.dataset) + '/full_lstm-bert_predictive_performances-OOD-' + str(
args.dataset) + '_ood2.json')
kuma_FullData = kuma_FullData[select_columns].iloc[0]
kuma_InDomain = kuma_InDomain[select_columns].iloc[0]
kuma_OOD1 = kuma_OOD1[select_columns].iloc[0]
kuma_OOD2 = kuma_OOD2[select_columns].iloc[0]
LSTM_FullData = LSTM_FullData[select_columns].iloc[0]
LSTM_InDomain = LSTM_InDomain[select_columns].iloc[0]
LSTM_OOD1 = LSTM_OOD1[select_columns].iloc[0]
LSTM_OOD2 = LSTM_OOD2[select_columns].iloc[0]
kuma_result = pd.concat([kuma_FullData, kuma_InDomain, kuma_OOD1, kuma_OOD2], ignore_index=False, axis=1).T
kuma_result = kuma_result.reset_index()[select_columns]
LSTM_result = pd.concat([LSTM_FullData, LSTM_InDomain, LSTM_OOD1, LSTM_OOD2], ignore_index=False, axis=1).T
LSTM_result = LSTM_result.reset_index()[select_columns]
if args.save_accuracy_version:
kuma_result = kuma_result.rename(columns={"mean-acc":"KUMA ACC", "std-acc":"KUMA std"})
LSTM_result = LSTM_result.rename(columns={"mean-acc":"LSTM ACC", "std-acc":"LSTM std"})
else:
kuma_result = kuma_result.rename(columns={"mean-f1":"KUMA F1", "std-f1":"KUMA std"})
LSTM_result = LSTM_result.rename(columns={"mean-f1":"LSTM F1", "std-f1":"LSTM std"})
def one_domain_len(domain):
overall = np.zeros(5)
for _j_, seed in enumerate([5,10,15,20,25]):
if 'ood' in str(domain):
path_to_file : str = f'kuma_model/{args.dataset}/kuma-bert-output_seed-kuma-bert{seed}-OOD-{args.dataset}_{domain}.npy'
elif 'full' in str(domain):
path_to_file : str = f'kuma_model/{args.dataset}_full/kuma-bert-output_seed-kuma-bert{seed}.npy'
else:
path_to_file : str = f'kuma_model/{args.dataset}/kuma-bert-output_seed-kuma-bert{seed}.npy'
file_data = np.load(path_to_file, allow_pickle=True).item()
aggregated_ratio = np.zeros(len(file_data))
for _i_, (docid, metadata) in enumerate(file_data.items()):
rationale_ratio = min(
1.,
metadata['rationale'].sum()/metadata['full text length']
)
aggregated_ratio[_i_] = rationale_ratio
overall[_j_] = aggregated_ratio.mean()
return overall.mean(), overall.std(), overall
Full_len_mean, Full_len_std, Full_len_overall = one_domain_len('full')
InDomain_len_mean, InDomain_len_std, InDomain_len_overall = one_domain_len('InDomain')
ood1_len_mean, ood1_len_std, ood1_len_overall= one_domain_len('ood1')
ood2_len_mean, ood2_len_std, ood2_len_overall= one_domain_len('ood2')
kuma_result['KUMA Len'] = [Full_len_mean,InDomain_len_mean,ood1_len_mean,ood2_len_mean]
#############################
SPECTRA = pd.read_csv('saved_everything/' + str(args.dataset) + '/spectra_mean.csv')[['avg', 'std']].rename(columns={"avg":"SPECTRA F1", "std":"SPECTRA std"})
##############################
final = pd.concat([bert_result, fresh_result, LSTM_result, kuma_result, SPECTRA], axis=1)
final['Domain'] = ['Full', 'SynD', 'AsyD1', 'AsyD2']
final = final.rename({'Domain': 'Testing Set'})
s = final[final.select_dtypes(exclude=['object']).columns] * 100
final[s.columns] = s
print(final.dtypes)
print(final)
if args.save_accuracy_version:
final.to_csv('saved_everything/' + str(args.dataset) + '/selective_results_acc.csv')
else:
final.to_csv('saved_everything/' + str(args.dataset) + '/selective_results.csv')